Missing Modalities Imputation via Cascaded Residual Autoencoder (CRA)

[Project page] [CVPR'17 Paper]

Library requirements

Caffe
(Matlab)

Data

Please create HDF5 files (see hdf4creation folder) with tags modal1, modal2, ...

Run the code

This is a part of our source code use for Multi-PIE experiment written in Caffe. The code is for a instance when the CRA is consisting of 3 RAs. Modification with different numbers of RA is straight-forward

Pretraining

pretrain3_CRA.sh

Finetunning

ft_CRA.sh

Citation

If you find this work useful, please cite our papers with the following bibtex:

@inproceedings{ tran2017missing,
  author = { Luan Tran and Xiaoming Liu and Jiayu Zhou and Rong Jin },
  title = { Missing Modalities Imputation via Cascaded Residual Autoencoder },
  booktitle = { In Proceeding of IEEE Computer Vision and Pattern Recognition },
  address = { Honolulu, HI },
  month = { July },
  year = { 2017 },
}

Contacts

If you have any questions, feel free to drop an email to tranluan@msu.edu.

sikid/CRA